Performing computations involving data stored in tables in a distributed computing system presents significant engineering challenges. A single request from a client might involve data stored in multiple table portions. Access to each table portion storing requested data may be controlled by various processes spread over multiple servers. Furthermore, the loads on the various processes and their respective servers may vary across the system, resulting in inefficiencies. Accordingly, it is desirable to provide an efficient manner of performing computations involving data stored in tables in a distributed computing system, while also performing load balancing.